1. Field of the Invention
The present invention relates to a learning method of an injection characteristic and to a fuel injection controller.
2. Description of Related Art
A known diesel engine performs a pilot injection of injecting fuel of a quantity less than a fuel quantity of a main injection before the main injection in order to inhibit a noise accompanying combustion or to improve an exhaust gas characteristic. Even if injection period command values, injection quantity command values (command injection quantities) and the like of the injectors are equalized in order to perform fuel injection control, a variation can be caused in the actually injected fuel quantities because of individual differences of the injectors. Specifically, the injection quantity of the pilot injection can become extremely small compared to the injection quantity of the main injection. Therefore, if a difference arises between the desired injection quantity and the actual injection quantity of the pilot injection, sufficient attainment of the above-mentioned objects will become difficult.
Therefore, a proposed controller, e.g., as described in JP-A-2003-254139, performs feedback control for conforming actual rotation speed of an engine to target rotation speed by performing N times of equally-divided fuel injections. The controller learns a learning value for compensating the difference between the command injection quantity and the desired injection quantity at that time. Furthermore, the controller performs the feedback control so that rotational fluctuation among the cylinders is compensated. The controller can grasp an injection characteristic at the time of performing a fuel injection of a minute quantity such as the pilot injection by performing the equally-divided N injections. As a result, an appropriate learning value can be obtained.
The time necessary for obtaining the learning value should be preferably as short as possible. However, when the processing for obtaining the learning value is performed for the first time, e.g., at the time of product shipment of the fuel injection controller, the time necessary for the feedback control to reach a stationary state tends to lengthen. If the learning is performed on the condition that a sufficient time for the first processing of obtaining the learning value elapses, obtainment of the learning value takes a long time. The inventor of the present invention found that accurate computation of a fluctuation correction value for compensating the rotation fluctuation among the cylinders becomes difficult if this time is shortened.
The difficulty in the learning of the variation in the injection characteristics among the cylinders with high accuracy while avoiding unnecessary lengthening of the time necessary for the learning is not limited to the learning of the pilot injection but is substantially common in fuel injection controllers compensating the variation in the injection characteristics among the cylinders.